The Virtual Machine Resource Allocation based on Service Features in Cloud Computing Environment

Transcription

1 Send Orders for Reprns o reprns@benhamscence.ae The Open Cybernecs & Sysemcs Journal, 2015, 9, Open Access The Vrual Machne Resource Allocaon based on Servce Feaures n Cloud Compung Envronmen Lu Yan and Guo Chenxan College of Informaon Engneerng, Zhongzhou Unversy, Zhengzhou, , P.R. Chna Absrac: Ths paper presens he vrual machne resource allocaon algorhm based on busness characerscs under a cloud compung envronmen. Ths algorhm descrbes he user servce characersc me delay facor and prce facor and oher parameers, resource scheduler compung daa cener accordng o he user on he busness characerscs of he descrpon and he saus of cloud calculaed he cos of he busness ndex. Accordng o he cos ndex dsrbuon of cloud compung resources, frsly he sysem model of he algorhm s nroduced, and hen he model s esablshed accordng o he correspondng problem of he sysem model, and hen he correspondng resource allocaon scheme s proposed. Fnally, he performance of he algorhm s verfed by smulaon. The smulaon resuls show ha he proposed algorhm n cloud compung plaform mproves he ulzaon rae of resources and a he same me sgnfcanly reduces he compuaonal characerscs of dfferen busness users of cloud use cos, and mproves he cloud user servce experence. Keywords: Busness characerscs, Cloud compung, Resource allocao Vrual machne. 1. INTRODUCTION Wh he developmen of cloud compung, more and more users wll do busness mgraon o he cloud servce provders of cloud compung fla sao a vrual machne runnng n he cloud compung plaform, and become he cloud users. Cloud servce provders perform accordng o he specfc needs of he cloud user busness dsrbuon of s correspondng vrual machne confguraon and quany. Usng he curren resource allocaon mehod can mprove he rae of cloud resources beer, and ncrease he cloud servce provder revenue. However, from he cloud user's pon of vew, hese mehods can no brng much benef o he cloud user, or even reduce he user experence, from a long-erm pon of vew; he long-erm ncome of a cloud servce provder wll also have a negave mpac. The busness characerscs of dfferen clouds of dfferen users vary from he vrual machne resource prce, delay and resource deploymen and oher requremens are also no dencal. Therefore, n he process of vrual machne resource allocao allocaon s necessary for vrual machne resources accordng o he characerscs of he cloud user, hereby reducng he cos of low cloud users, and mprovng he cloud user servce experence, and hen realzng maxmum cloud servce provders of long-erm ncome [1]. In hs paper, we from he cloud user perspecve proposed he vrual machne allocaon algorhm based on a cloud compung busness characerscs, n order o mprove he ulzaon rae of he plaform n cloud compung resources a he same Address correspondence o hs auhor a he College of Informaon Engneerng, Zhongzhou Unversy, Zhengzhou, Hena , P.R. Chna; Tel: ; E-mal: luyan0407@126.com X/15 me reducng he cloud user cos, and mprovng he cloud user servce experence. 2. THE SYSTEM MODEL Consderng a compung plaform s dsrbued n a pluraly of dfferen geographc locaon daa ceners o form clouds. Capacy of each daa cener, delay performance and vrual machne resource prces are no he same. Cloud servce provders perform accordng o he user's requremens for he allocaon of cloud resources correspondng vrual machne and operaon of cloud users of vrual machne n he daa cener [2]. In order o undersand, followng defnons are gven: Defnon one: a resource scheduler, runnng sae of compung daa cener accordng o he user's requremens and he cloud vrual machne resource allocaon correspondng o he cloud user, a daa cener resource scheduler self runs n cloud compung plaform. Defnon wo: daa cener saus monorng process, runnng on each daa cener, runnng sae real-me monorng daa cener, he amoun of resdual resources ncludng daa cener and vrual machne resource prce nformao saus nformaon and o he resource scheduler sends locaon daa cener. Defnon hree: daa cener saus able, daa sorage cener sae nformaon for resource scheduler module, accordng o he sae nformaon cener daa process sae monorng repors are updaed n real me, ncludng he daa cener of he resdual amoun of resources, me delay and resource prce nformaon. Accordng o he above defno he sysem archecure can be represened as n Fg. (1) Benham Open

2 640 The Open Cybernecs & Sysemcs Journal, 2015, Volume 9 Yan and Chenxan Fg. (1). Vrual machne sysem framework of dsrbuon maps based on busness characerscs. ( ) ( ) In Fg. (1), R l, n,c represens he cloud user submed o he resource scheduler resource reques, whereas l = l f, m, h, b represens a cloud user for vrual machne confgurao ncludng he processor f, memory m, exernal memory bandwdh h and b. To represen a number of cloud users, s needed for vrual machne; c represens busness characerscs of he cloud user ; L, P, D respecvely represen he amoun of remanng resources daa cener and n repors daa cener saus monorng process o a resource scheduler, resource prce and me delay esmaon nformao whereas, he resource prce P = P( f,m,h,b ), ncludng processor compung resource prce f, resource n memory m, exernal sor- age resource h and bandwdh resource b. In order o smplfy he sysem model, we assume a cloud user wh only one ype of busness [3]. 3. MODEL CONSTRUCTION AND PROBLEM DE- SCRIPTION 3.1. Descrpon of Servce Feaures For he specfc busness, such as massvely mulplayer onlne game busness, he busness of large amoun of nernal nformaon neraco vrual machne deploymen for he use of cenralzed deploymen mode, and hghly sensve o he delay, resource prce has lle nfluence on. As mcro-blog socal busness, wh srong regonal characerscs, he smaller amoun of nernal busness nformaon neracon s more suable for vrual machne deploymen adopng dsrbued deploymen mode, and a he same me, delay n hs knd of busness generaes low sensvy of prce. The dfference of busness of cloud users leads o s varable busness characerscs, and accordng o he vrual machne deploymen requremens, me delay facor and prce facor can be used o descrbe he busness characerscs of cloud user. c = ( s,a,b ) (1) 3.2. Saus Updaes of Daa Cener On he prcng model of resource prce daa cener and cloud provder, n he sysem model, akng no accoun he equalzaon of mulple daa cener load demand, we used dynamc prcng model based on he resdual amoun of resources. Specfcally, daa cener resources prce P was k, calculaed by he daa cener saus monorng process accordng o he followng formula. P k, = g 1 P k,o +g 2 f L k, ( ) (2) In he formula, k represens he daa cenre number, represen he curren me, g as weghng facor, 1, P k,o s a reference prce of resources daa cener, k s consa f ( L k, ) s he funcon assocaed wh he resdual amoun of resources. Tme delay esmaon of daa cener renewal process: resource scheduler perodcally Xang Yun compung daa cener sendng delay es sgnal, and record he response me delay d he es sgnal, sendng perodc esng sgnal for k,, and whenever a resource scheduler receves he sae d nformaon from he daa cener wll also send o he daa cener of an expermen he es sgnal. Delay of daa cener s obaned by he formula: D d I k, k, (3) 1 I where I represens he mos recen I sgnal, namely he mean response delay me delay I recenly a es sgnal o he daa cener esmaon Descrpon of he Problem In order o nfluence he nuve descrpon of daa cener delay and resource prce on he busness, we pu forward

3 The Vrual Machne Resource Allocaon based on Servce Feaures The Open Cybernecs & Sysemcs Journal, 2015, Volume he concep of busness cos ndex, resource scheduler accordng o he cloud user servce cos ndex for he allocaon of resources o he correspondng busness. Cos ndex can be calculaed by he followng formula: T = a D + b P (4), k, k,, k, Among hem, T represens he cloud user arrved a, k, momen of s busness wh respec o daa cener k cos ndex, P, k, = Pk, l represens he user n he daa cener, k s deployed on a vrual machne ha confgured he requremens of cos, he same cloud users of s busness delay facor and prce facor are consan. From he ype se, on he same cloud user servce, dfferen daa ceners and dfferen me, s cos ndex are no he same. Accordng o busness characerscs of cloud users assgn vrual machne resource, mnmum cloud user cos, cos ndex mnmzaon problem s equvalen o solvng a cloud user, herefore, our objecve funcon can be descrbed as follows: K mn T T x,, k,, k, k 1 K, k, (5) k 1 s.. x n x, k, l Lk,, k 1,2,..., K Among hem, T, represens he oal cos of he cloud user, represens he cloud user servce reques arrval x represens he vrual machne allocaon ndex, me,, k, where users of vrual machne n k daa cener are he upper deparmen number. The frs resrcon represens a number of vrual machne resources allocaed o he vrual machne scheduler oal user should be equal o he user requred; second lmng condons ndcaes ha he daa cener on k allocaed o he vrual machne resources requred of user should be less han he oal amoun of resources avalable o he daa cener k. Known by he frs lmng condons, he objecve funcon s equvalen o: m n x T n T (6) K, k,,, k, k 1 n In he formula, for a parcular user servce, he number of vrual machne needs n s deermned, and he cos of ndex T user servce relave o a parcular daa cener s, k, he fxed value. Therefore, we can solve he waer fllng algorhm hrough he opmzaon problem [4]. The seps are as follows: 1) Accordng o he cos ndex o sor he daa cener, remember he ranked se for he D T vrual machne, he nalzaon allocaon ndex: x,k, = 0; 2) Selec he daa cener, cos ndex T mnmum D, k, T f x l L, hen daa cener for he user servce agency, a, k, k, vrual machne, x = x +, and more new n = n - 1 1, k,, k, and L = L - x l, f x l L k, k,, k,, k, k, execue sep 1); >, hen D = D - k, 3) DT = f, ndcaes ha he curren cloud compung plaform does no have suffcen resources o provde servces, for he user allocaon falure, end; f x,k = n, and he dsrbuon end, oherwse execue sep 2); Through many mes of erave waer fllng algorhm, we can fnd he bes vrual machne user I deploymen sraegy. 4. RESOURCE ALLOCATION ALGORITHM BASED ON BUSINESS CHARACTERISTICS Accordng o he sysem model and he analyss of he problem, we propose a vrual machne resource allocaon algorhm based on busness characerscs. The specfc process s shown n Fg. (2). The concree seps of he proposed vrual machne allocaon algorhm are as follows: Sep one, he user o subm a reques for a resource allo- R l, n, c ; cang resources Sep wo, f he user's vrual machne deploymen sraegy of s 1, execue sep hree; f s v, v 1, hen execue sep four; Sep hree, cenralzed way for user o he vrual machne resource allocao he specfc mehod for: 1) Resource scheduler queres he daa cener k saus able, eleced o mee he needs of he user resources daa cener R ( l, n ), mus sasfy L k ³ n, l, and consues he canddae daa cener se G of user. 2) If he canddae user se daa cener G s null, suggess ha he curren cloud compung plaform has no enough resources allocaed o he user I, resource scheduler refused o user reques, execue sep fve; oherwse, he calculaon of canddae daa cener cos ndex T k Î G T T, k,, relave cloud users I collecon of G n each daa cener; 3) Selec he canddae daa cener G collecon cos ndex and he lowes n T,k, daa cener k for user deparmen cloud vrual machne; 4) Remanng resources sae of L on he seleced daa k, cener are updaed, execue sep fve; Sep four, adopng dsrbued deploymen of vrual machne as he user, he specfc mehod for:

4 642 The Open Cybernecs & Sysemcs Journal, 2015, Volume 9 Yan and Chenxan Fg. (2). Flow char of algorhm. 1) Resource scheduler queres he daa cener saus able, R l, n o mee he user selec he resource requremen ( ) daa cener he surplus resources, n l, where L s k, n represens he number of user k, mus sasfy Lk, needed for vrual machne, l represens he cloud user desred vrual machne confgurao v represens he user requremens of vrual machne whch are deployed n v daa cener, he daa cener of a canddae daa cener user n he se G G ; 2) If he canddae daa cener G collecon of daa cener < v number, represens ha he curren daa are no adequae o mee he cloud cener user resource requremens, resource scheduler refused o user requess for resources, execue sep fve; oherwse, he calculaon of he canddae se of each daa cener n G phase o he cloud user cos of ndex T,,, k Î G s performed; k n 3) In he canddae se G conssng of a collecon of daa cener cos ndex T, k,, k mnmum v daa cener s seleced G collecon of he daa cener are user vr- G n he ual machne deploymen. 4) The remanng resource sae L on he seleced daa k, cener are updaed, execue sep fve; Sep fve, as he vrual machne assgnmen complees, he resource scheduler o he cloud user I feedback resuls. Compared wh he waer fllng algorhm sandards, he above algorhm consderng he sraegy requres he user

5 The Vrual Machne Resource Allocaon based on Servce Feaures The Open Cybernecs & Sysemcs Journal, 2015, Volume for he vrual machne deploymen wh some performance loss, bu sll sgnfcanly reduces he user's cloud compung cos, and especally requres he use of cenralzed vrual machne deploymen sraegy users for beer performance [5]. 5. SIMULATION AND PERFORMANCE ANALYSIS 5.1. Esablshmen of he Smulaon Model In pracce, he daa cener wh delay and prce resource ulzaon rae, herefore, we frs defne he ulzaon daa cener resources h, momen daa cener k resources ulzaon rae calculaon formula s as follows: h k, = m 1 f k, + m 2 m k, + m 3 h k, + m 4 b k, m 1 f k + m 2 m k + m 3 h k + m 4 b k (7) In he formula, m represens he correspondng o he daa cener resources nfluence facor, delay of 0 1, 1 ; he denomnaor ndcaes he oal amoun of resources daa cener k, he denomnaor represens he use of a correspondng amoun of resources n he me. Accordng o he ulzaon rae of daa cener resources, we defne he relaonshp beween delay and resource ulzaon rae n he smulaon model as follows: é 1 ù D k, = D k,0 expê - eú ëê 1-h k, ûú In he formula, D reference delay daa cener k, accordng o he formula, when he resource ulzaon rae s k,0 less han 70%, he resource ulzaon rae ncreases slowly; when he resource ulzaon rae s hgher han 70%, he ncrease of resource ulzaon rae ncreases rapdly [6]. Smlarly, we defne he rae relaonshp by he prce of resource and resource daa cener n he smulaon model as follows: 1 P k, = P 1-h k,o + P k,o (9) k, Among hem, P represen he benchmark prce of resources daa cener k, whch can be seen from he above k, o equaon; wh he ncrease of resource ulzao resource prce daa cener also gradually ncreases The Resuls of Smulaon Analyss Consder a dsrbued cloud compung plaform conssng of fve dfferen geographc locaon daa formed n he cener of each daa cener, all havng he same capacy, bu s reference resources prce and reference delay are no he same. In each me slo, he busness needs a number of vrual machnes and he correspondng busness characerscs randomly generaed and execued n a daa cener busness leavng a ceran probably. The specfc smulaon parameers are shown n Table 1. (8) Table 1: The smulaon parameers. Parameers Value The number of daa cener 5 Daa cener compung ably Daa cener memory capacy The bandwdh of daa cener MIPS 16GBye 100Mbps Leave he probably daa cener operaons 0.02 Busness average requred number of vrual machne 3 The number of smulaon me slo 288 Fgs. (3) and (4), respecvely, ndcae he resource prce fve daa cener (Fg. 3) and delay (Fg. 4) n he 100h me slo o slo changes among he 200h. In he smulaon model, he load s proporonal o he prce of resources and me delay and daa ceners,.e. he larger he load, he hgher he prce of resources, and he delay s large. Fve daa ceners benchmark prce and he reference delay are no he same. From he daa cener I o V, he benchmark prce lowers and he reference delay s ncreased. Fg. (5) shows he fve daa ceners n he 100h o 200h me slo resource ulzaon curves, as can be seen from he graph. Resource allocaon algorhm, fve daa cener load dfference, and he prce model algorhm wh balanced load effec are proposed n hs paper. Accordng o he smulaon model, he average load of fve daa cener s 60%, whle he load n Fg. (5) daa cener flucuaes beween 40%-70%, demonsrang ha he proposed algorhm can acheve beer load balancng daa cener beween he arge [7]. Fg. (6) shows he relaon graph of rae and delay, resource prce daa cener ulzaon III resources, as can be seen from he graph of me delay and resource prces and resource ulzaon rae s proporonal o, and has a hgh correlaon conssen wh he smulaon model heory, llusrang he correcness of smulaon [8]. We concluded facor a more han 0.7 busness as delay sensve raffc, and he prce facor b more han 0.7 busness as prce sensve busness. Boh were sascally dfferen vrual machne deploymen sraeges of delay sensve raffc delay and resource cos and prce sensve busness accumulaed value, Fg. (7). Among hem, he frs acs as dsrbued vrual machne deploymen delay and prce dfferen characerscs under he busness sraegy. I can be observed from he fgure ha for delay sensve raffc, he cumulave delay value s far less han he prce sensve busness Fg. (7a). Ths s because of he busness offce delay sensve busness requremens of smaller physcal and me delay, and prce sensve busness o busness processng delay s no a src requremen; on he conrary, he me delay sensve servce resource cos accumulaon s greaer han he prce sensve servce, Fg. (7b). Ths s because he prce sensve busness requremens of resource prce are as low as possble, whle he prce of delay sensve busness o resource s no src; Fg. (7c) and (7d) show cenralzed vrual machne deploymen delay and

6 644 The Open Cybernecs & Sysemcs Journal, 2015, Volume 9 Yan and Chenxan Fg. (3). Resource prce curves of fve daa cener. Fg. (4). Tme delay curve of fve daa cener. Fg. (5). The ulzaon rae of daa cener resources.

7 The Vrual Machne Resource Allocaon based on Servce Feaures The Open Cybernecs & Sysemcs Journal, 2015, Volume Fg. (6). Dagram of he relaonshp beween resources ulzaon and rae, me delay n daa cener III(DCIII). Fg. (7a). Tme delay of dfferen servce under dsrbued VM deploymen sraeges. Fg. (7b). Resource cos of dfferen servce under dsrbued VM deploymen sraeges.

8 646 The Open Cybernecs & Sysemcs Journal, 2015, Volume 9 Yan and Chenxan Fg. (7c). Tme delay of dfferen servce under cenralzed VM deploymen sraeges. Fg. (7d). Resource cos of dfferen servce under cenralzed VM deploymen sraeges. Fg. (7). Cos and me delay curve of he dfferen characerscs busness under dfferen vrual machne deploymen sraeges. resource prce busness sraeges under he dfferen characerscs of he cumulave curve; he resul s smlar o Fgs. (7a) and (7b), whch shows ha no maer wha he user vrual machne deploymen sraegy s, he proposed algorhm can be calculaed usng he correspondng cos reduce users of cloud. The cenralzed vrual machne deploymen sraegy users, dfferences beween dfferen ypes of raffc delays and resource prces are more obvous, he resuls show ha he proposed algorhm s beer o adop for cenralzed vrual machne deploymen sraegy users reurn. In general, Fg. (7) shows ha he vrual machne resource allocaon algorhm s based on busness characerscs ha can allocae resources for users accordng o busness characerscs of users, hus reducng he cos of he user [9, 10]. CONCLUSION From he above-menoned cone resource allocaon algorhm based on busness characerscs manly nroduces a cloud compung envronmen. Ths algorhm can perform accordng o he characerscs of user servce o allocae approprae resources for users, hus reducng he user cos, load balance and can acheve a cloud compung plaform. Ths paper frs nroduces he sysem model of he algorhm, laer he model s esablshed accordng o he correspondng problem of he sysem model, and hen he correspondng resource allocaon scheme s proposed. Fnally, he performance of he algorhm s verfed by smulaon. The smulaon resuls show ha he proposed algorhm can perform accordng o he characerscs of user servce

9 The Vrual Machne Resource Allocaon based on Servce Feaures The Open Cybernecs & Sysemcs Journal, 2015, Volume for a parcular resource allocaon o users, sgnfcanly reducng he cos o he user, and mplemenng a dsrbued cloud daa cener compung plaform under dfferen load balancng. CONFLICT OF INTEREST The auhors confrm ha hs arcle conen has no conflc of neres. ACKNOWLEDGEMENTS Ths work was fnancally suppored by he Laonng Docor Sar-up Foundaon ( ). REFERENCES [1] L. Xu W. Chen and Z. Wang, A sraegy of dynamc resource schedulng n cloud daa cener, In: Proc. of IEEE Cluser Workshops, pp , Sep [2] V. Jalapar, and G. D. Nguye Cloud resource allocaon games, Dep of Compuer Scence, Tech. Rep, Dec [3] Kuln Che C-RAN Whe Paper, verson 2.5, Chna Moble Research Insue, Oc [4] L. Guangje, Z. Senje and Y. Xueb Archecure of GPP based, scalable, large-scale C-RAN BBU pool, In: Proc. of IEEE GC Workshops, Dec. 2012, pp [5] Z. B. Zhu, P. Gupa, and Q. Wang, Vrual base saon pool: owards a wreless nework cloud for rado access neworks, In: Proc. of he 8 h ACM Inernaonal Conference on Compung Froners Aprl 2011, p. 34. [6] G. Bhanage, I. Seskar, and R. Mahndra, Vrual basesaon: archecure for an open shared WMAX framework, In: Proc. of he 2 nd ACM SIGCOMM Workshop on Vrualzed Infrasrucure Sysems and Archecures, Sep. 2010, pp [7] S. Namba, T. Warab and, S. Kaneko, BBU-RRH swchng schemes for cenralzed RAN, In: Proc. of IEEE CHINACOM, Aug. 2012, pp [8] S. Bhaumk, S. P. Chandrabose, and M. K. Jaaprolu, CloudIQ: a framework for processng base saons n a daa cener, In: Proc. of he 18 h Annual Inernaonal Conference on Moble Compung and Neworkng, Aug. 2012, pp [9] X. F. Tao, Y. Z. Hou, and K. D. Wang, GPP-based sof base saon desgnng and opmzao Journal of Compuer Scence and Technology, vol. 28, no. 3, pp , [10] Y. Che X. L and F. Che Overvew and analyss of cloud compung research and applcao In: Proc. of IEEE ICEE, May 2011, pp Receved: Aprl 02, 2015 Revsed: May 23, 2015 Acceped: June 06, 2015 Yan and Chenxan; Lcensee Benham Open. Ths s an open access arcle lcensed under he erms of he Creave Commons Arbuon Non-Commercal Lcense (hp://creavecommons.org/- lcenses/by-nc/3.0/) whch perms unresrced, non-commercal use, dsrbuon and reproducon n any medum, provded he work s properly ced.